1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
5 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
6 #include <linux/libnvdimm.h>
7 #include <linux/sched/mm.h>
8 #include <linux/vmalloc.h>
9 #include <linux/uaccess.h>
10 #include <linux/module.h>
11 #include <linux/blkdev.h>
12 #include <linux/fcntl.h>
13 #include <linux/async.h>
14 #include <linux/genhd.h>
15 #include <linux/ndctl.h>
16 #include <linux/sched.h>
17 #include <linux/slab.h>
18 #include <linux/cpu.h>
28 static int nvdimm_bus_major;
29 struct class *nd_class;
30 static DEFINE_IDA(nd_ida);
32 static int to_nd_device_type(struct device *dev)
35 return ND_DEVICE_DIMM;
36 else if (is_memory(dev))
37 return ND_DEVICE_REGION_PMEM;
38 else if (is_nd_blk(dev))
39 return ND_DEVICE_REGION_BLK;
40 else if (is_nd_dax(dev))
41 return ND_DEVICE_DAX_PMEM;
42 else if (is_nd_region(dev->parent))
43 return nd_region_to_nstype(to_nd_region(dev->parent));
48 static int nvdimm_bus_uevent(struct device *dev, struct kobj_uevent_env *env)
50 return add_uevent_var(env, "MODALIAS=" ND_DEVICE_MODALIAS_FMT,
51 to_nd_device_type(dev));
54 static struct module *to_bus_provider(struct device *dev)
56 /* pin bus providers while regions are enabled */
57 if (is_nd_region(dev)) {
58 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
60 return nvdimm_bus->nd_desc->module;
65 static void nvdimm_bus_probe_start(struct nvdimm_bus *nvdimm_bus)
67 nvdimm_bus_lock(&nvdimm_bus->dev);
68 nvdimm_bus->probe_active++;
69 nvdimm_bus_unlock(&nvdimm_bus->dev);
72 static void nvdimm_bus_probe_end(struct nvdimm_bus *nvdimm_bus)
74 nvdimm_bus_lock(&nvdimm_bus->dev);
75 if (--nvdimm_bus->probe_active == 0)
76 wake_up(&nvdimm_bus->wait);
77 nvdimm_bus_unlock(&nvdimm_bus->dev);
80 static int nvdimm_bus_probe(struct device *dev)
82 struct nd_device_driver *nd_drv = to_nd_device_driver(dev->driver);
83 struct module *provider = to_bus_provider(dev);
84 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
87 if (!try_module_get(provider))
90 dev_dbg(&nvdimm_bus->dev, "START: %s.probe(%s)\n",
91 dev->driver->name, dev_name(dev));
93 nvdimm_bus_probe_start(nvdimm_bus);
94 debug_nvdimm_lock(dev);
95 rc = nd_drv->probe(dev);
96 debug_nvdimm_unlock(dev);
99 nd_region_probe_success(nvdimm_bus, dev);
101 nd_region_disable(nvdimm_bus, dev);
102 nvdimm_bus_probe_end(nvdimm_bus);
104 dev_dbg(&nvdimm_bus->dev, "END: %s.probe(%s) = %d\n", dev->driver->name,
108 module_put(provider);
112 static int nvdimm_bus_remove(struct device *dev)
114 struct nd_device_driver *nd_drv = to_nd_device_driver(dev->driver);
115 struct module *provider = to_bus_provider(dev);
116 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
119 if (nd_drv->remove) {
120 debug_nvdimm_lock(dev);
121 rc = nd_drv->remove(dev);
122 debug_nvdimm_unlock(dev);
124 nd_region_disable(nvdimm_bus, dev);
126 dev_dbg(&nvdimm_bus->dev, "%s.remove(%s) = %d\n", dev->driver->name,
128 module_put(provider);
132 static void nvdimm_bus_shutdown(struct device *dev)
134 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
135 struct nd_device_driver *nd_drv = NULL;
138 nd_drv = to_nd_device_driver(dev->driver);
140 if (nd_drv && nd_drv->shutdown) {
141 nd_drv->shutdown(dev);
142 dev_dbg(&nvdimm_bus->dev, "%s.shutdown(%s)\n",
143 dev->driver->name, dev_name(dev));
147 void nd_device_notify(struct device *dev, enum nvdimm_event event)
151 struct nd_device_driver *nd_drv;
153 nd_drv = to_nd_device_driver(dev->driver);
155 nd_drv->notify(dev, event);
157 nd_device_unlock(dev);
159 EXPORT_SYMBOL(nd_device_notify);
161 void nvdimm_region_notify(struct nd_region *nd_region, enum nvdimm_event event)
163 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
168 /* caller is responsible for holding a reference on the device */
169 nd_device_notify(&nd_region->dev, event);
171 EXPORT_SYMBOL_GPL(nvdimm_region_notify);
173 struct clear_badblocks_context {
174 resource_size_t phys, cleared;
177 static int nvdimm_clear_badblocks_region(struct device *dev, void *data)
179 struct clear_badblocks_context *ctx = data;
180 struct nd_region *nd_region;
181 resource_size_t ndr_end;
184 /* make sure device is a region */
185 if (!is_nd_pmem(dev))
188 nd_region = to_nd_region(dev);
189 ndr_end = nd_region->ndr_start + nd_region->ndr_size - 1;
191 /* make sure we are in the region */
192 if (ctx->phys < nd_region->ndr_start
193 || (ctx->phys + ctx->cleared) > ndr_end)
196 sector = (ctx->phys - nd_region->ndr_start) / 512;
197 badblocks_clear(&nd_region->bb, sector, ctx->cleared / 512);
199 if (nd_region->bb_state)
200 sysfs_notify_dirent(nd_region->bb_state);
205 static void nvdimm_clear_badblocks_regions(struct nvdimm_bus *nvdimm_bus,
206 phys_addr_t phys, u64 cleared)
208 struct clear_badblocks_context ctx = {
213 device_for_each_child(&nvdimm_bus->dev, &ctx,
214 nvdimm_clear_badblocks_region);
217 static void nvdimm_account_cleared_poison(struct nvdimm_bus *nvdimm_bus,
218 phys_addr_t phys, u64 cleared)
221 badrange_forget(&nvdimm_bus->badrange, phys, cleared);
223 if (cleared > 0 && cleared / 512)
224 nvdimm_clear_badblocks_regions(nvdimm_bus, phys, cleared);
227 long nvdimm_clear_poison(struct device *dev, phys_addr_t phys,
230 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
231 struct nvdimm_bus_descriptor *nd_desc;
232 struct nd_cmd_clear_error clear_err;
233 struct nd_cmd_ars_cap ars_cap;
234 u32 clear_err_unit, mask;
235 unsigned int noio_flag;
241 nd_desc = nvdimm_bus->nd_desc;
243 * if ndctl does not exist, it's PMEM_LEGACY and
244 * we want to just pretend everything is handled.
249 memset(&ars_cap, 0, sizeof(ars_cap));
250 ars_cap.address = phys;
251 ars_cap.length = len;
252 noio_flag = memalloc_noio_save();
253 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, &ars_cap,
254 sizeof(ars_cap), &cmd_rc);
255 memalloc_noio_restore(noio_flag);
260 clear_err_unit = ars_cap.clear_err_unit;
261 if (!clear_err_unit || !is_power_of_2(clear_err_unit))
264 mask = clear_err_unit - 1;
265 if ((phys | len) & mask)
267 memset(&clear_err, 0, sizeof(clear_err));
268 clear_err.address = phys;
269 clear_err.length = len;
270 noio_flag = memalloc_noio_save();
271 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_CLEAR_ERROR, &clear_err,
272 sizeof(clear_err), &cmd_rc);
273 memalloc_noio_restore(noio_flag);
279 nvdimm_account_cleared_poison(nvdimm_bus, phys, clear_err.cleared);
281 return clear_err.cleared;
283 EXPORT_SYMBOL_GPL(nvdimm_clear_poison);
285 static int nvdimm_bus_match(struct device *dev, struct device_driver *drv);
287 static struct bus_type nvdimm_bus_type = {
289 .uevent = nvdimm_bus_uevent,
290 .match = nvdimm_bus_match,
291 .probe = nvdimm_bus_probe,
292 .remove = nvdimm_bus_remove,
293 .shutdown = nvdimm_bus_shutdown,
296 static void nvdimm_bus_release(struct device *dev)
298 struct nvdimm_bus *nvdimm_bus;
300 nvdimm_bus = container_of(dev, struct nvdimm_bus, dev);
301 ida_simple_remove(&nd_ida, nvdimm_bus->id);
305 bool is_nvdimm_bus(struct device *dev)
307 return dev->release == nvdimm_bus_release;
310 struct nvdimm_bus *walk_to_nvdimm_bus(struct device *nd_dev)
314 for (dev = nd_dev; dev; dev = dev->parent)
315 if (is_nvdimm_bus(dev))
317 dev_WARN_ONCE(nd_dev, !dev, "invalid dev, not on nd bus\n");
319 return to_nvdimm_bus(dev);
323 struct nvdimm_bus *to_nvdimm_bus(struct device *dev)
325 struct nvdimm_bus *nvdimm_bus;
327 nvdimm_bus = container_of(dev, struct nvdimm_bus, dev);
328 WARN_ON(!is_nvdimm_bus(dev));
331 EXPORT_SYMBOL_GPL(to_nvdimm_bus);
333 struct nvdimm_bus *nvdimm_to_bus(struct nvdimm *nvdimm)
335 return to_nvdimm_bus(nvdimm->dev.parent);
337 EXPORT_SYMBOL_GPL(nvdimm_to_bus);
339 struct nvdimm_bus *nvdimm_bus_register(struct device *parent,
340 struct nvdimm_bus_descriptor *nd_desc)
342 struct nvdimm_bus *nvdimm_bus;
345 nvdimm_bus = kzalloc(sizeof(*nvdimm_bus), GFP_KERNEL);
348 INIT_LIST_HEAD(&nvdimm_bus->list);
349 INIT_LIST_HEAD(&nvdimm_bus->mapping_list);
350 init_waitqueue_head(&nvdimm_bus->wait);
351 nvdimm_bus->id = ida_simple_get(&nd_ida, 0, 0, GFP_KERNEL);
352 if (nvdimm_bus->id < 0) {
356 mutex_init(&nvdimm_bus->reconfig_mutex);
357 badrange_init(&nvdimm_bus->badrange);
358 nvdimm_bus->nd_desc = nd_desc;
359 nvdimm_bus->dev.parent = parent;
360 nvdimm_bus->dev.release = nvdimm_bus_release;
361 nvdimm_bus->dev.groups = nd_desc->attr_groups;
362 nvdimm_bus->dev.bus = &nvdimm_bus_type;
363 nvdimm_bus->dev.of_node = nd_desc->of_node;
364 dev_set_name(&nvdimm_bus->dev, "ndbus%d", nvdimm_bus->id);
365 rc = device_register(&nvdimm_bus->dev);
367 dev_dbg(&nvdimm_bus->dev, "registration failed: %d\n", rc);
373 put_device(&nvdimm_bus->dev);
376 EXPORT_SYMBOL_GPL(nvdimm_bus_register);
378 void nvdimm_bus_unregister(struct nvdimm_bus *nvdimm_bus)
382 device_unregister(&nvdimm_bus->dev);
384 EXPORT_SYMBOL_GPL(nvdimm_bus_unregister);
386 static int child_unregister(struct device *dev, void *data)
389 * the singular ndctl class device per bus needs to be
390 * "device_destroy"ed, so skip it here
392 * i.e. remove classless children
397 if (is_nvdimm(dev)) {
398 struct nvdimm *nvdimm = to_nvdimm(dev);
399 bool dev_put = false;
401 /* We are shutting down. Make state frozen artificially. */
402 nvdimm_bus_lock(dev);
403 nvdimm->sec.state = NVDIMM_SECURITY_FROZEN;
404 if (test_and_clear_bit(NDD_WORK_PENDING, &nvdimm->flags))
406 nvdimm_bus_unlock(dev);
407 cancel_delayed_work_sync(&nvdimm->dwork);
411 nd_device_unregister(dev, ND_SYNC);
416 static void free_badrange_list(struct list_head *badrange_list)
418 struct badrange_entry *bre, *next;
420 list_for_each_entry_safe(bre, next, badrange_list, list) {
421 list_del(&bre->list);
424 list_del_init(badrange_list);
427 static int nd_bus_remove(struct device *dev)
429 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
431 mutex_lock(&nvdimm_bus_list_mutex);
432 list_del_init(&nvdimm_bus->list);
433 mutex_unlock(&nvdimm_bus_list_mutex);
435 wait_event(nvdimm_bus->wait,
436 atomic_read(&nvdimm_bus->ioctl_active) == 0);
439 device_for_each_child(&nvdimm_bus->dev, NULL, child_unregister);
441 spin_lock(&nvdimm_bus->badrange.lock);
442 free_badrange_list(&nvdimm_bus->badrange.list);
443 spin_unlock(&nvdimm_bus->badrange.lock);
445 nvdimm_bus_destroy_ndctl(nvdimm_bus);
450 static int nd_bus_probe(struct device *dev)
452 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
455 rc = nvdimm_bus_create_ndctl(nvdimm_bus);
459 mutex_lock(&nvdimm_bus_list_mutex);
460 list_add_tail(&nvdimm_bus->list, &nvdimm_bus_list);
461 mutex_unlock(&nvdimm_bus_list_mutex);
463 /* enable bus provider attributes to look up their local context */
464 dev_set_drvdata(dev, nvdimm_bus->nd_desc);
469 static struct nd_device_driver nd_bus_driver = {
470 .probe = nd_bus_probe,
471 .remove = nd_bus_remove,
474 .suppress_bind_attrs = true,
475 .bus = &nvdimm_bus_type,
476 .owner = THIS_MODULE,
477 .mod_name = KBUILD_MODNAME,
481 static int nvdimm_bus_match(struct device *dev, struct device_driver *drv)
483 struct nd_device_driver *nd_drv = to_nd_device_driver(drv);
485 if (is_nvdimm_bus(dev) && nd_drv == &nd_bus_driver)
488 return !!test_bit(to_nd_device_type(dev), &nd_drv->type);
491 static ASYNC_DOMAIN_EXCLUSIVE(nd_async_domain);
493 void nd_synchronize(void)
495 async_synchronize_full_domain(&nd_async_domain);
497 EXPORT_SYMBOL_GPL(nd_synchronize);
499 static void nd_async_device_register(void *d, async_cookie_t cookie)
501 struct device *dev = d;
503 if (device_add(dev) != 0) {
504 dev_err(dev, "%s: failed\n", __func__);
509 put_device(dev->parent);
512 static void nd_async_device_unregister(void *d, async_cookie_t cookie)
514 struct device *dev = d;
516 /* flush bus operations before delete */
517 nvdimm_bus_lock(dev);
518 nvdimm_bus_unlock(dev);
520 device_unregister(dev);
524 void __nd_device_register(struct device *dev)
530 * Ensure that region devices always have their NUMA node set as
531 * early as possible. This way we are able to make certain that
532 * any memory associated with the creation and the creation
533 * itself of the region is associated with the correct node.
535 if (is_nd_region(dev))
536 set_dev_node(dev, to_nd_region(dev)->numa_node);
538 dev->bus = &nvdimm_bus_type;
540 get_device(dev->parent);
541 if (dev_to_node(dev) == NUMA_NO_NODE)
542 set_dev_node(dev, dev_to_node(dev->parent));
546 async_schedule_dev_domain(nd_async_device_register, dev,
550 void nd_device_register(struct device *dev)
552 device_initialize(dev);
553 __nd_device_register(dev);
555 EXPORT_SYMBOL(nd_device_register);
557 void nd_device_unregister(struct device *dev, enum nd_async_mode mode)
564 * In the async case this is being triggered with the
565 * device lock held and the unregistration work needs to
566 * be moved out of line iff this is thread has won the
567 * race to schedule the deletion.
569 if (!kill_device(dev))
573 async_schedule_domain(nd_async_device_unregister, dev,
578 * In the sync case the device is being unregistered due
579 * to a state change of the parent. Claim the kill state
580 * to synchronize against other unregistration requests,
581 * or otherwise let the async path handle it if the
582 * unregistration was already queued.
585 killed = kill_device(dev);
586 nd_device_unlock(dev);
592 device_unregister(dev);
596 EXPORT_SYMBOL(nd_device_unregister);
599 * __nd_driver_register() - register a region or a namespace driver
600 * @nd_drv: driver to register
601 * @owner: automatically set by nd_driver_register() macro
602 * @mod_name: automatically set by nd_driver_register() macro
604 int __nd_driver_register(struct nd_device_driver *nd_drv, struct module *owner,
605 const char *mod_name)
607 struct device_driver *drv = &nd_drv->drv;
610 pr_debug("driver type bitmask not set (%ps)\n",
611 __builtin_return_address(0));
615 if (!nd_drv->probe) {
616 pr_debug("%s ->probe() must be specified\n", mod_name);
620 drv->bus = &nvdimm_bus_type;
622 drv->mod_name = mod_name;
624 return driver_register(drv);
626 EXPORT_SYMBOL(__nd_driver_register);
628 int nvdimm_revalidate_disk(struct gendisk *disk)
630 struct device *dev = disk_to_dev(disk)->parent;
631 struct nd_region *nd_region = to_nd_region(dev->parent);
632 int disk_ro = get_disk_ro(disk);
635 * Upgrade to read-only if the region is read-only preserve as
636 * read-only if the disk is already read-only.
638 if (disk_ro || nd_region->ro == disk_ro)
641 dev_info(dev, "%s read-only, marking %s read-only\n",
642 dev_name(&nd_region->dev), disk->disk_name);
643 set_disk_ro(disk, 1);
648 EXPORT_SYMBOL(nvdimm_revalidate_disk);
650 static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
653 return sprintf(buf, ND_DEVICE_MODALIAS_FMT "\n",
654 to_nd_device_type(dev));
656 static DEVICE_ATTR_RO(modalias);
658 static ssize_t devtype_show(struct device *dev, struct device_attribute *attr,
661 return sprintf(buf, "%s\n", dev->type->name);
663 static DEVICE_ATTR_RO(devtype);
665 static struct attribute *nd_device_attributes[] = {
666 &dev_attr_modalias.attr,
667 &dev_attr_devtype.attr,
672 * nd_device_attribute_group - generic attributes for all devices on an nd bus
674 struct attribute_group nd_device_attribute_group = {
675 .attrs = nd_device_attributes,
677 EXPORT_SYMBOL_GPL(nd_device_attribute_group);
679 static ssize_t numa_node_show(struct device *dev,
680 struct device_attribute *attr, char *buf)
682 return sprintf(buf, "%d\n", dev_to_node(dev));
684 static DEVICE_ATTR_RO(numa_node);
686 static struct attribute *nd_numa_attributes[] = {
687 &dev_attr_numa_node.attr,
691 static umode_t nd_numa_attr_visible(struct kobject *kobj, struct attribute *a,
694 if (!IS_ENABLED(CONFIG_NUMA))
701 * nd_numa_attribute_group - NUMA attributes for all devices on an nd bus
703 struct attribute_group nd_numa_attribute_group = {
704 .attrs = nd_numa_attributes,
705 .is_visible = nd_numa_attr_visible,
707 EXPORT_SYMBOL_GPL(nd_numa_attribute_group);
709 int nvdimm_bus_create_ndctl(struct nvdimm_bus *nvdimm_bus)
711 dev_t devt = MKDEV(nvdimm_bus_major, nvdimm_bus->id);
714 dev = device_create(nd_class, &nvdimm_bus->dev, devt, nvdimm_bus,
715 "ndctl%d", nvdimm_bus->id);
718 dev_dbg(&nvdimm_bus->dev, "failed to register ndctl%d: %ld\n",
719 nvdimm_bus->id, PTR_ERR(dev));
720 return PTR_ERR_OR_ZERO(dev);
723 void nvdimm_bus_destroy_ndctl(struct nvdimm_bus *nvdimm_bus)
725 device_destroy(nd_class, MKDEV(nvdimm_bus_major, nvdimm_bus->id));
728 static const struct nd_cmd_desc __nd_cmd_dimm_descs[] = {
729 [ND_CMD_IMPLEMENTED] = { },
732 .out_sizes = { 4, 128, },
734 [ND_CMD_SMART_THRESHOLD] = {
736 .out_sizes = { 4, 8, },
738 [ND_CMD_DIMM_FLAGS] = {
740 .out_sizes = { 4, 4 },
742 [ND_CMD_GET_CONFIG_SIZE] = {
744 .out_sizes = { 4, 4, 4, },
746 [ND_CMD_GET_CONFIG_DATA] = {
748 .in_sizes = { 4, 4, },
750 .out_sizes = { 4, UINT_MAX, },
752 [ND_CMD_SET_CONFIG_DATA] = {
754 .in_sizes = { 4, 4, UINT_MAX, },
760 .in_sizes = { 4, 4, UINT_MAX, },
762 .out_sizes = { 4, 4, UINT_MAX, },
766 .in_sizes = { sizeof(struct nd_cmd_pkg), UINT_MAX, },
768 .out_sizes = { UINT_MAX, },
772 const struct nd_cmd_desc *nd_cmd_dimm_desc(int cmd)
774 if (cmd < ARRAY_SIZE(__nd_cmd_dimm_descs))
775 return &__nd_cmd_dimm_descs[cmd];
778 EXPORT_SYMBOL_GPL(nd_cmd_dimm_desc);
780 static const struct nd_cmd_desc __nd_cmd_bus_descs[] = {
781 [ND_CMD_IMPLEMENTED] = { },
784 .in_sizes = { 8, 8, },
786 .out_sizes = { 4, 4, 4, 4, },
788 [ND_CMD_ARS_START] = {
790 .in_sizes = { 8, 8, 2, 1, 5, },
792 .out_sizes = { 4, 4, },
794 [ND_CMD_ARS_STATUS] = {
796 .out_sizes = { 4, 4, UINT_MAX, },
798 [ND_CMD_CLEAR_ERROR] = {
800 .in_sizes = { 8, 8, },
802 .out_sizes = { 4, 4, 8, },
806 .in_sizes = { sizeof(struct nd_cmd_pkg), UINT_MAX, },
808 .out_sizes = { UINT_MAX, },
812 const struct nd_cmd_desc *nd_cmd_bus_desc(int cmd)
814 if (cmd < ARRAY_SIZE(__nd_cmd_bus_descs))
815 return &__nd_cmd_bus_descs[cmd];
818 EXPORT_SYMBOL_GPL(nd_cmd_bus_desc);
820 u32 nd_cmd_in_size(struct nvdimm *nvdimm, int cmd,
821 const struct nd_cmd_desc *desc, int idx, void *buf)
823 if (idx >= desc->in_num)
826 if (desc->in_sizes[idx] < UINT_MAX)
827 return desc->in_sizes[idx];
829 if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA && idx == 2) {
830 struct nd_cmd_set_config_hdr *hdr = buf;
832 return hdr->in_length;
833 } else if (nvdimm && cmd == ND_CMD_VENDOR && idx == 2) {
834 struct nd_cmd_vendor_hdr *hdr = buf;
836 return hdr->in_length;
837 } else if (cmd == ND_CMD_CALL) {
838 struct nd_cmd_pkg *pkg = buf;
840 return pkg->nd_size_in;
845 EXPORT_SYMBOL_GPL(nd_cmd_in_size);
847 u32 nd_cmd_out_size(struct nvdimm *nvdimm, int cmd,
848 const struct nd_cmd_desc *desc, int idx, const u32 *in_field,
849 const u32 *out_field, unsigned long remainder)
851 if (idx >= desc->out_num)
854 if (desc->out_sizes[idx] < UINT_MAX)
855 return desc->out_sizes[idx];
857 if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA && idx == 1)
859 else if (nvdimm && cmd == ND_CMD_VENDOR && idx == 2)
861 else if (!nvdimm && cmd == ND_CMD_ARS_STATUS && idx == 2) {
863 * Per table 9-276 ARS Data in ACPI 6.1, out_field[1] is
864 * "Size of Output Buffer in bytes, including this
867 if (out_field[1] < 4)
870 * ACPI 6.1 is ambiguous if 'status' is included in the
871 * output size. If we encounter an output size that
872 * overshoots the remainder by 4 bytes, assume it was
873 * including 'status'.
875 if (out_field[1] - 4 == remainder)
877 return out_field[1] - 8;
878 } else if (cmd == ND_CMD_CALL) {
879 struct nd_cmd_pkg *pkg = (struct nd_cmd_pkg *) in_field;
881 return pkg->nd_size_out;
887 EXPORT_SYMBOL_GPL(nd_cmd_out_size);
889 void wait_nvdimm_bus_probe_idle(struct device *dev)
891 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
894 if (nvdimm_bus->probe_active == 0)
896 nvdimm_bus_unlock(dev);
897 nd_device_unlock(dev);
898 wait_event(nvdimm_bus->wait,
899 nvdimm_bus->probe_active == 0);
901 nvdimm_bus_lock(dev);
905 static int nd_pmem_forget_poison_check(struct device *dev, void *data)
907 struct nd_cmd_clear_error *clear_err =
908 (struct nd_cmd_clear_error *)data;
909 struct nd_btt *nd_btt = is_nd_btt(dev) ? to_nd_btt(dev) : NULL;
910 struct nd_pfn *nd_pfn = is_nd_pfn(dev) ? to_nd_pfn(dev) : NULL;
911 struct nd_dax *nd_dax = is_nd_dax(dev) ? to_nd_dax(dev) : NULL;
912 struct nd_namespace_common *ndns = NULL;
913 struct nd_namespace_io *nsio;
914 resource_size_t offset = 0, end_trunc = 0, start, end, pstart, pend;
916 if (nd_dax || !dev->driver)
919 start = clear_err->address;
920 end = clear_err->address + clear_err->cleared - 1;
922 if (nd_btt || nd_pfn || nd_dax) {
928 ndns = nd_dax->nd_pfn.ndns;
935 nsio = to_nd_namespace_io(&ndns->dev);
936 pstart = nsio->res.start + offset;
937 pend = nsio->res.end - end_trunc;
939 if ((pstart >= start) && (pend <= end))
946 static int nd_ns_forget_poison_check(struct device *dev, void *data)
948 return device_for_each_child(dev, data, nd_pmem_forget_poison_check);
951 /* set_config requires an idle interleave set */
952 static int nd_cmd_clear_to_send(struct nvdimm_bus *nvdimm_bus,
953 struct nvdimm *nvdimm, unsigned int cmd, void *data)
955 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
957 /* ask the bus provider if it would like to block this request */
958 if (nd_desc->clear_to_send) {
959 int rc = nd_desc->clear_to_send(nd_desc, nvdimm, cmd, data);
965 /* require clear error to go through the pmem driver */
966 if (!nvdimm && cmd == ND_CMD_CLEAR_ERROR)
967 return device_for_each_child(&nvdimm_bus->dev, data,
968 nd_ns_forget_poison_check);
970 if (!nvdimm || cmd != ND_CMD_SET_CONFIG_DATA)
973 /* prevent label manipulation while the kernel owns label updates */
974 wait_nvdimm_bus_probe_idle(&nvdimm_bus->dev);
975 if (atomic_read(&nvdimm->busy))
980 static int __nd_ioctl(struct nvdimm_bus *nvdimm_bus, struct nvdimm *nvdimm,
981 int read_only, unsigned int ioctl_cmd, unsigned long arg)
983 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
984 const struct nd_cmd_desc *desc = NULL;
985 unsigned int cmd = _IOC_NR(ioctl_cmd);
986 struct device *dev = &nvdimm_bus->dev;
987 void __user *p = (void __user *) arg;
988 char *out_env = NULL, *in_env = NULL;
989 const char *cmd_name, *dimm_name;
990 u32 in_len = 0, out_len = 0;
991 unsigned int func = cmd;
992 unsigned long cmd_mask;
993 struct nd_cmd_pkg pkg;
999 desc = nd_cmd_dimm_desc(cmd);
1000 cmd_name = nvdimm_cmd_name(cmd);
1001 cmd_mask = nvdimm->cmd_mask;
1002 dimm_name = dev_name(&nvdimm->dev);
1004 desc = nd_cmd_bus_desc(cmd);
1005 cmd_name = nvdimm_bus_cmd_name(cmd);
1006 cmd_mask = nd_desc->cmd_mask;
1010 if (cmd == ND_CMD_CALL) {
1011 if (copy_from_user(&pkg, p, sizeof(pkg)))
1015 if (!desc || (desc->out_num + desc->in_num == 0) ||
1016 !test_bit(cmd, &cmd_mask))
1019 /* fail write commands (when read-only) */
1023 case ND_CMD_SET_CONFIG_DATA:
1024 case ND_CMD_ARS_START:
1025 case ND_CMD_CLEAR_ERROR:
1027 dev_dbg(dev, "'%s' command while read-only.\n",
1028 nvdimm ? nvdimm_cmd_name(cmd)
1029 : nvdimm_bus_cmd_name(cmd));
1035 /* process an input envelope */
1036 in_env = kzalloc(ND_CMD_MAX_ENVELOPE, GFP_KERNEL);
1039 for (i = 0; i < desc->in_num; i++) {
1042 in_size = nd_cmd_in_size(nvdimm, cmd, desc, i, in_env);
1043 if (in_size == UINT_MAX) {
1044 dev_err(dev, "%s:%s unknown input size cmd: %s field: %d\n",
1045 __func__, dimm_name, cmd_name, i);
1049 if (in_len < ND_CMD_MAX_ENVELOPE)
1050 copy = min_t(u32, ND_CMD_MAX_ENVELOPE - in_len, in_size);
1053 if (copy && copy_from_user(&in_env[in_len], p + in_len, copy)) {
1060 if (cmd == ND_CMD_CALL) {
1061 func = pkg.nd_command;
1062 dev_dbg(dev, "%s, idx: %llu, in: %u, out: %u, len %llu\n",
1063 dimm_name, pkg.nd_command,
1064 in_len, out_len, buf_len);
1067 /* process an output envelope */
1068 out_env = kzalloc(ND_CMD_MAX_ENVELOPE, GFP_KERNEL);
1074 for (i = 0; i < desc->out_num; i++) {
1075 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i,
1076 (u32 *) in_env, (u32 *) out_env, 0);
1079 if (out_size == UINT_MAX) {
1080 dev_dbg(dev, "%s unknown output size cmd: %s field: %d\n",
1081 dimm_name, cmd_name, i);
1085 if (out_len < ND_CMD_MAX_ENVELOPE)
1086 copy = min_t(u32, ND_CMD_MAX_ENVELOPE - out_len, out_size);
1089 if (copy && copy_from_user(&out_env[out_len],
1090 p + in_len + out_len, copy)) {
1094 out_len += out_size;
1097 buf_len = (u64) out_len + (u64) in_len;
1098 if (buf_len > ND_IOCTL_MAX_BUFLEN) {
1099 dev_dbg(dev, "%s cmd: %s buf_len: %llu > %d\n", dimm_name,
1100 cmd_name, buf_len, ND_IOCTL_MAX_BUFLEN);
1105 buf = vmalloc(buf_len);
1111 if (copy_from_user(buf, p, buf_len)) {
1116 nd_device_lock(dev);
1117 nvdimm_bus_lock(dev);
1118 rc = nd_cmd_clear_to_send(nvdimm_bus, nvdimm, func, buf);
1122 rc = nd_desc->ndctl(nd_desc, nvdimm, cmd, buf, buf_len, &cmd_rc);
1126 if (!nvdimm && cmd == ND_CMD_CLEAR_ERROR && cmd_rc >= 0) {
1127 struct nd_cmd_clear_error *clear_err = buf;
1129 nvdimm_account_cleared_poison(nvdimm_bus, clear_err->address,
1130 clear_err->cleared);
1133 if (copy_to_user(p, buf, buf_len))
1137 nvdimm_bus_unlock(dev);
1138 nd_device_unlock(dev);
1146 enum nd_ioctl_mode {
1151 static int match_dimm(struct device *dev, void *data)
1153 long id = (long) data;
1155 if (is_nvdimm(dev)) {
1156 struct nvdimm *nvdimm = to_nvdimm(dev);
1158 return nvdimm->id == id;
1164 static long nd_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
1165 enum nd_ioctl_mode mode)
1168 struct nvdimm_bus *nvdimm_bus, *found = NULL;
1169 long id = (long) file->private_data;
1170 struct nvdimm *nvdimm = NULL;
1173 ro = ((file->f_flags & O_ACCMODE) == O_RDONLY);
1174 mutex_lock(&nvdimm_bus_list_mutex);
1175 list_for_each_entry(nvdimm_bus, &nvdimm_bus_list, list) {
1176 if (mode == DIMM_IOCTL) {
1179 dev = device_find_child(&nvdimm_bus->dev,
1180 file->private_data, match_dimm);
1183 nvdimm = to_nvdimm(dev);
1185 } else if (nvdimm_bus->id == id) {
1190 atomic_inc(&nvdimm_bus->ioctl_active);
1194 mutex_unlock(&nvdimm_bus_list_mutex);
1200 rc = __nd_ioctl(nvdimm_bus, nvdimm, ro, cmd, arg);
1203 put_device(&nvdimm->dev);
1204 if (atomic_dec_and_test(&nvdimm_bus->ioctl_active))
1205 wake_up(&nvdimm_bus->wait);
1210 static long bus_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1212 return nd_ioctl(file, cmd, arg, BUS_IOCTL);
1215 static long dimm_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1217 return nd_ioctl(file, cmd, arg, DIMM_IOCTL);
1220 static int nd_open(struct inode *inode, struct file *file)
1222 long minor = iminor(inode);
1224 file->private_data = (void *) minor;
1228 static const struct file_operations nvdimm_bus_fops = {
1229 .owner = THIS_MODULE,
1231 .unlocked_ioctl = bus_ioctl,
1232 .compat_ioctl = bus_ioctl,
1233 .llseek = noop_llseek,
1236 static const struct file_operations nvdimm_fops = {
1237 .owner = THIS_MODULE,
1239 .unlocked_ioctl = dimm_ioctl,
1240 .compat_ioctl = dimm_ioctl,
1241 .llseek = noop_llseek,
1244 int __init nvdimm_bus_init(void)
1248 rc = bus_register(&nvdimm_bus_type);
1252 rc = register_chrdev(0, "ndctl", &nvdimm_bus_fops);
1254 goto err_bus_chrdev;
1255 nvdimm_bus_major = rc;
1257 rc = register_chrdev(0, "dimmctl", &nvdimm_fops);
1259 goto err_dimm_chrdev;
1262 nd_class = class_create(THIS_MODULE, "nd");
1263 if (IS_ERR(nd_class)) {
1264 rc = PTR_ERR(nd_class);
1268 rc = driver_register(&nd_bus_driver.drv);
1275 class_destroy(nd_class);
1277 unregister_chrdev(nvdimm_major, "dimmctl");
1279 unregister_chrdev(nvdimm_bus_major, "ndctl");
1281 bus_unregister(&nvdimm_bus_type);
1286 void nvdimm_bus_exit(void)
1288 driver_unregister(&nd_bus_driver.drv);
1289 class_destroy(nd_class);
1290 unregister_chrdev(nvdimm_bus_major, "ndctl");
1291 unregister_chrdev(nvdimm_major, "dimmctl");
1292 bus_unregister(&nvdimm_bus_type);
1293 ida_destroy(&nd_ida);